From: Amir Shehata <ashehata@whamcloud.com>
Date: Tue, 21 Aug 2018 19:23:26 +0000 (-0700)
Subject: LU-11272 lnet: router handling
X-Git-Tag: 2.11.55~23
X-Git-Url: https://git.whamcloud.com/?a=commitdiff_plain;h=05becd69bc0c79fde00f0fddf4935ed8d8e3beb3;p=fs%2Flustre-release.git

LU-11272 lnet: router handling

Re-create the md and mdh if the router checker ping times out.
When re-transmitting a message do so even if the peer is marked down
to fulfill the message's retry quota.

Test-Parameters: trivial
Signed-off-by: Amir Shehata <ashehata@whamcloud.com>
Change-Id: I7b2a1ec6602dac9a112f4d318b0512f68f923969
Reviewed-on: https://review.whamcloud.com/33043
Tested-by: Jenkins
Tested-by: Maloo <hpdd-maloo@intel.com>
Reviewed-by: Olaf Weber <olaf.weber@hpe.com>
Reviewed-by: Sonia Sharma <sharmaso@whamcloud.com>
Reviewed-by: Oleg Drokin <green@whamcloud.com>
---

diff --git a/lnet/lnet/lib-move.c b/lnet/lnet/lib-move.c
index a89fa41..eee5901 100644
--- a/lnet/lnet/lib-move.c
+++ b/lnet/lnet/lib-move.c
@@ -865,7 +865,8 @@ lnet_peer_is_alive(struct lnet_peer_ni *lp, time64_t now)
 /* NB: returns 1 when alive, 0 when dead, negative when error;
  *     may drop the lnet_net_lock */
 static int
-lnet_peer_alive_locked(struct lnet_ni *ni, struct lnet_peer_ni *lp)
+lnet_peer_alive_locked(struct lnet_ni *ni, struct lnet_peer_ni *lp,
+		       struct lnet_msg *msg)
 {
 	time64_t now = ktime_get_seconds();
 
@@ -876,6 +877,13 @@ lnet_peer_alive_locked(struct lnet_ni *ni, struct lnet_peer_ni *lp)
 		return 1;
 
 	/*
+	 * If we're resending a message, let's attempt to send it even if
+	 * the peer is down to fulfill our resend quota on the message
+	 */
+	if (msg->msg_retry_count > 0)
+		return 1;
+
+	/*
 	 * Peer appears dead, but we should avoid frequent NI queries (at
 	 * most once per lnet_queryinterval seconds).
 	 */
@@ -933,7 +941,7 @@ lnet_post_send_locked(struct lnet_msg *msg, int do_send)
 
 	/* NB 'lp' is always the next hop */
 	if ((msg->msg_target.pid & LNET_PID_USERFLAG) == 0 &&
-	    lnet_peer_alive_locked(ni, lp) == 0) {
+	    lnet_peer_alive_locked(ni, lp, msg) == 0) {
 		the_lnet.ln_counters[cpt]->drop_count++;
 		the_lnet.ln_counters[cpt]->drop_length += msg->msg_len;
 		lnet_net_unlock(cpt);
diff --git a/lnet/lnet/router.c b/lnet/lnet/router.c
index 14d3cae..dc80e40 100644
--- a/lnet/lnet/router.c
+++ b/lnet/lnet/router.c
@@ -1070,7 +1070,14 @@ lnet_ping_router_locked(struct lnet_peer_ni *rtr)
 	}
 
 	rcd = rtr->lpni_rcd;
-	if (!rcd || rcd->rcd_nnis > rcd->rcd_pingbuffer->pb_nnis)
+
+	/*
+	 * The response to the router checker ping could've timed out and
+	 * the mdh might've been invalidated, so we need to update it
+	 * again.
+	 */
+	if (!rcd || rcd->rcd_nnis > rcd->rcd_pingbuffer->pb_nnis ||
+	    LNetMDHandleIsInvalid(rcd->rcd_mdh))
 		rcd = lnet_update_rc_data_locked(rtr);
 	if (rcd == NULL)
 		return;